Import

The performance of an import mode dataset is largely driven by fundamental design decisions, such as the granularity of fact and dimension tables. For example, large dimension tables with more than a million unique values, such as customer IDs or product IDs will produce much less performant report queries than small dimensions with only 100 to 1,000 unique values. Likewise, DAX Measures that access columns containing thousands of unique values will perform much more slowly than measures that reference columns with a few unique values. A simplistic but effective understanding is that higher levels of cardinality (unique values) result in greater memory consumption via reduced compression and CPUs require additional time to scan greater amounts of memory. 

An import mode designer should be cautious about the performance implications of relationships to large dimension tables. Although usability is somewhat compromised, a separate but less granular dimension containing only the most common columns can be created to drive more efficient report queries. For example, business users may rarely need to access individual product SKUs and would prefer the performance benefit provided by a smaller dimension table that contains only product categories and product subcategories.